[FieldTrip] defining phase and axial to planar conversion

[jan-mathijs schoffelen]

Hi Sangita,

> I'd like to convert axial gradiometer data to planar data as part of a time frequency analysis.  I'm currently using the 'megplanar' code with the 'sincos' option, then applying time frequency analysis on the dH and dV components separately, and then recombining the resulting power spectra with the 'combineplanar' script.  This all is working well.

Good.

> I'd also like to do the same time frequency analysis described above using the phase of a relatively low frequency oscillation (i.e. theta or alpha troughs) as the time locking events within each sensor.  What are your opinions on how to best do this?
> 
> My current procedure: 
> 1. Apply megplanar to get the dH and dV components
> 2.  Identify phase time points of interest in the dH and dV  components separately (i.e. band pass filter and then determine analytic phase of dH and dV components separately)
> 3. Perform two separate time frequency analyses time locked to the phase points of interest in the dH and dV components separately
> 
> Is there a better way to do this?  
> 
> If this method seems ok, does summing the dH and dV  power spectra after step 3 seem justified even if the phase troughs extracted from the dH and dV components and used for time locking are not comparable on a 1:1 basis?  Or would it make more sense to do two separate analyses for dH and dV and never recombine?

I wouldn't do the dH and dV analysis separately, and then combine. One of the reasons is indeed that it would be strange if there is a phase shift between the dV and dH gradients. Another reason is that the actual direction of dH and dV is rather arbitrary. One could consider an oscillatory source to yield a time-varying gradient oriented exactly along the dH-axis, exactly along the dV-axis, but also somewhere in between. In the first case, the source of interest is not visible along the dV-axis (and the phases estimated from the dV are actually rubbish, so introduces a lot of non-specific variance on your feature of interest), in the second case the situation is reversed (dH picks up nothing relevant), and in the in between situation both pick up a bit. What I would do is to do a PCA on each dH-dV combination, and then go for the first PCA-component that will be used for the phase computation.
This approach is implemented in ft_combineplanar, but only for frequency domain data. The order of the steps would be:

cfg = [];
cfg.planarmethod = 'sincos';
dataplanar = ft_megplanar(cfg, data);

cfg = [];
cfg.method = 'mtmconvol'/'mtmfft'
cfg.output = 'fourier';
cfg.taper = 'hanning';
freqplanar = ft_freqanalysis(cfg, dataplanar);

cfg = [];
cfg.method = 'svd'; % which is equivalent to PCA
freqcombined = ft_combineplanar(cfg, freqplanar);


freqplanar.fourierspctrm should now contain phase information, estimated at the gradient orientation yielding the maximum power (assuming a fixed orientation over time and trials, but different across frequencies).

Alternatively, you can do the analysis in the time domain, but this will require some code of your own.

Best wishes,

Jan-Mathijs

> 
> Thanks in advance for any help!
> Sangita
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip

Jan-Mathijs Schoffelen, MD PhD 

Donders Institute for Brain, Cognition and Behaviour, 
Centre for Cognitive Neuroimaging,
Radboud University Nijmegen, The Netherlands

Max Planck Institute for Psycholinguistics,
Nijmegen, The Netherlands

J.Schoffelen at donders.ru.nl
Telephone: +31-24-3614793

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20130118/170a2252/attachment.html>